Visualizing Data

Just what the world needs — another blog.

Well, when it comes to the sharing the best practices for displaying healthcare data visually and finding and telling the story buried in your data that is EXACTLY what the world needs — a blog that delivers the information and help you've just got to have, but don't have easy access to.

And as much as I love the sound of my own voice (and I do, ask anyone) I encourage you to contribute your thoughts, questions and examples (HIPAA compliant please — I don't look good in stripes).

Let the blogging begin.

Bar Humbug!

Can I tell you how tired I am of hearing people tell me that “bars are boring”? VERY TIRED. Bar humbug!! A true pillar of data visualization, bars are used to display and compare values in health and healthcare data — values like number, frequency, distribution, or other measures (e.g. a statistical mean or median) for different discrete categories of data. Bars help us order data by ranking, or see and understand the distribution of a dataset, such as whether it is skewed to the left or right, or is normally distributed. We can use bars to show deviation (difference or changes) from a baseline or benchmark, and they’re especially useful for displaying and comparing relative differences in data with wide-ranging absolute values. Bars are highly adaptable and may be arranged either horizontally or vertically depending on overall layout, space constraints, and labeling requirements. When combined with other media such as points and lines, they help us see and understand important contextual information — to consider the “compared to what” and “so what?” questions essential to all great data analysis and data visualizations. Bars are NOT boring. Only the unimaginative use of them is. Here are some possibilities for using bars effectively. DISTRIBUTIONS Bars are the go-to graph to show how often (number of times) something is observed|occurs (the frequency distribution) in a dataset. Remember that they may be oriented horizontally or vertically depending on factors such as labels and space on a page; and that the base of the bar must always start at zero on the X or Y axis scale to show the entirety of the value, and avoid exaggerating small differences. A type of bar chart called a histogram is most commonly used to show the frequency distribution of data, organized into “bins,” of consecutive, non-overlapping intervals of a variable. Histograms are very useful in displaying the distribution of continuous intervals of data (ages, days, time, etc.), and determining whether the data is distributed relatively evenly, is skewed, or takes some other interesting shape, as in the following examples: RANKING Displaying values in bar charts in rank order affords us a simple and elegant way to convey relative differences in health and healthcare data, such as “more vs. fewer people (per 1,000 population) diagnosed with HIV in different countries,” or “hospitals that had the highest vs. lowest rate of hospital-acquired infections,” or “which communities had the lowest vaccination rates.” CHANGE OVER TIME Time series data may be displayed using bars if it’s important to see the values in direct comparison to one another. Additionally, and because we often encounter seasonality in health and healthcare data, displaying it over time using a bar chart may be preferable for some results, as it helps us see not only chronological change, but also the distribution of results. Sometimes, due to space constraints, or the number of time periods that need to be displayed, a horizontal bar chart is the best choice. PROPORTIONS: PART-TO-WHOLE & DIFFERENT CATEGORIES OF DATA Small Multiples Bar Chart is a helpful technique when we have three or more parts of the whole to display. By creating a series of bar charts, all with the same axes and scale, to show and compare different parts of the whole, we can make it far easier to see and compare those multiple parts. We are also able to view and compare directly the category of data being displayed in each column. Bar charts also extraordinarily well designed to display different categories of data across multiple dimensions. DEVIATION (DIFFERENCE, VARIATION) A bar graph may also be used to display a deviation; that is, how one or more sets of quantitative values differ from a reference set of values. They’re especially helpful when we need to compare differences or changes between groups that have a wide range of absolute values, such as departmental budgets vs. actual results, or different countries’ spending on healthcare services in one year and another. By using a deviation bar graph to display the relative differences, we can quickly identify which values are up or down, larger or smaller, and which ones may require further inquiry and analysis. Displaying results in this way tells viewers how far over or under target or goal they are — that is, the real difference from target, their actual score displayed on the X or Y axis, and performance compared to other groups or in other time periods. RANGES AND COMPARATIVE VALUES A floating bar chart may be used to display the range of a category of data (minimum and maximum value, beginning and ending values such as start and stop times, comparative data values such as percentile results). Some examples of how we’ve used these floating bars include displaying operational metrics such as hospital surgical case starts and stop times in multiple operating rooms; patient visit times in a clinic office by the day of the week and time of day; or comparison percentile ranges for performance metrics like patient experience survey results. When we need to display a wide range of values (500 to 10,000, say), we can place two bar charts with different axis ranges next to each other. One bar chart shows only the lower values with the longer bars truncated, while the other shows the full range of values. This simple technique allows us to display the smallest values and the largest values and makes clear to the viewer that the range is large. This technique also follows the best practice of starting both bar graphs at zero. Bars truly are the workhorses of data visualization. Once you have a solid understanding of the fundamental ways they can be used, you will acquire much greater skill and confidence as you create a range of simple to complex displays. I’m confident that the more you work with them, the more you’ll come to agree that bars are anything but boring — rather, used correctly and imaginatively, they are both gratifying and beautiful in their ability to help us see and understand the important stories in our health and healthcare data. Here’s wishing you a happy holiday season — and no more “Bar Humbug!”
Posted in Communicating Data to the Public, Data Visualization, Graphs, Newsletters | Leave a comment

Heavenly Icons

It’s quite astonishing how even when I get out of my office and resolve not think about my work for a while, I stumble upon something that makes me think about… my work. (Oh, the sweet and eternally elusive fantasy of escape.)

Take this morning. Bret and I were on a walk in the woods with our pups when we happened upon an old cemetery and this antique gravestone:

Finding an old cemetery in the woods is interesting, but it was the relief on the gravestone of a finger pointing toward the heavens that really caught my attention. Was this an instruction for those who would be helping the deceased (John Young) into his next life, or simply a message left for all of us earthbound souls about where he had gone — a reassurance of sorts? No matter: it amused me immensely and made me think about the directional icons we use in data visualizations.

Dead people, data viz icons… don’t even try to imagine living in my head.

We sometimes use icons in our data visualizations to grab a viewer’s attention and direct it toward something important or that requires action. Icons alert the viewer to “see this,” or say in clear, firm tones, “this is urgent; do something!”

The icon shapes we use most often are up- and down-facing carets to show that one measure is greater or less then another (for example, actual performance compared to target|budget), or — in the case of a performance over time — the way something is trending.

Depending upon what result or direction is desired, color may be applied to the icons: red for an undesirable value, say, and green for a desirable one. A word of caution here: you have to think through the use of these types of icons very carefully. If different measures with different directionality of good v. bad are displayed together, the mix may cause confusion for all viewers.

And yes, I know that we discourage the use of the colors red and green together, because about 8% of the world’s population is red|green color-blind; but if it makes sense to use this type of icon for those who are not, go ahead. The 8% can still perceive the directionality conveyed by the carets.

Keep in mind, of course, the above-noted problem of mixing metrics with different performance goals. (If this stuff were simple, everyone would do it right without our help!)

If we don’t want or need to show directionality or greater than|less than specifically, we simply use points and different saturations of a single color (usually red) to signal degrees of difference between one measure and another — to tell the viewer where to look first, beginning with the deepest saturation, then moving to the next and the next.

And yes, dear reader, it’s true that red|green color-blind people can’t see the red color on these icons, but they can perceive a shape that signals “look here.” They can see saturations of color from deepest (most urgent) to lightest (relatively less urgent) that display something like this:

We do NOT, however, use green and red points together for several reasons. First (and yes, I know this may be provocative in some circles, which makes it even more fun for me to write about), the purpose of a true management dashboard isn’t to pat anyone on the back with lots of green points that signal “good job!”; rather, we need to keep our displays uncluttered and use this type of icon only to call out and highlight areas of concern (you know: the type of poor performance that might sink the ship or land your organization on the front page of the local newspaper — and not in a good way).

[View full-size graphic.]

Another example of how we have used icons to communicate and draw viewers’ attention to a specific performance in comparison to percentiles is shown in the following display of State Medicaid agency results on Behavioral Health Care metrics. Here, we’ve used both shape and color to convey information.

A blue checkmark indicates performance at or above the 75th percentile. The purple minus sign in a circle for performance falling between the 25th and 75th percentile is somewhat neutral. The “Hey! Look here: this needs immediate attention” 25th-percentile-or-below results icon is a red x.

[View the interactive viz.]

There are also occasions when you should not use any color conveying good or bad performance on an icon. For example, you simply want to communicate that a value has increased or decreased. The viewer’s knowledge of the metric is required to determine whether that change is a good or a bad thing.

The following illustrates what I mean. The arrow icons show only that a hospital payor type has increased or decreased from one year to the next. Given the complexity of hospital contracting and reimbursement, we don’t have enough information to decide what is good or bad for this hospital, so we don’t use a signifying color. We leave that judgment to the contracts and reimbursement experts who work there.

[View the full dashboard.]

After seeing John Young’s gravestone, I got to wondering about some of the other icons that exist in our day to day world. Is the up arrow in the image below pointing to the heavens, or merely showing me the way to my meeting on the 8th floor? Let’s hope it’s the latter!


Posted in Communicating Data to the Public, Dashboards, Data Visualization, Newsletters | Leave a comment

2019 Reading (and Listening) List for the Health and Healthcare Data Geek

I’m heading to Bustins Island, Maine, in a few days to swing in the hammock and sip watermelon mojitos — Nirvana. I’m looking forward to this summer’s trek even more than usual because leading data visualization expert (and my mentor and good friend) Steve Few will be joining me and my family there for a few days.

I’m so excited! I get to share my family’s favorite summer spot with an important person in my life, and he’s stuck on an island with me, so I can talk to him to my heart’s content (if you’re reading this, Steve, it’s not too late to run, LOL)!

It seems only fitting, therefore, that the first title on my 2019 Summer Reading List is Steve’s most recent book.

Book Recommendations

Big Data, Big Dupe: A little book about a big bunch of nonsense, by Stephen Few

Here’s an excerpt from Steve’s website about this book:

“Big Data, Big Dupe is a little book about a big bunch of nonsense… While others have written about the dangers of Big Data, Stephen Few reveals the deceit that belies its illusory nature. If “data is the new oil,” Big Data is the new snake oil. It isn’t real. It’s a marketing campaign that has distracted us for years from the real and important work of deriving value from data.”

As I told Steve, when I first learned that he was writing this book, I didn’t think I would be very interested in it. But I did read it, and as always his words compelled me to stop and think about my work using data very deeply. The warning bells that Steve rang most loudly for me signaled these points:

  • Big data as one or more concept[s] remains ill-defined.
  • There is no clear or empirical reason to accept big data advocates’ claims that the sheer volume of data allows us to abandon 21st-century statistical data science methodology and methods; in fact, doing so would be both foolhardy and dangerous.
  • Abandonment of a scientific approach to analyzing data would jeopardize the true, necessary promise of quantitative, evidence-driven decision-making.

This last point is especially sobering given the tremendous amount of work required to transform our health and healthcare system. Few’s book is a thought-provoking, quick and easy read that I heartily recommend.

Dopesick: Dealers, Doctors, and the Drug Company that Addicted America, by Beth Macy

There has been no shortage of news coverage about the opioid epidemic in the U.S., so much so that I admit to being numbed by it at times. I am therefore especially grateful for Macy’s book. She uses first-person interviews to look squarely at the lives of young heroin users and their long-suffering parents, the drug-dealers and the cops, and the judges, doctors, and health activists struggling to fight the epidemic.

The author also details the actions of pharmaceutical company executives who aggressively marketed opioids, and uses data to help us to understand these people’s and companies’ truly disturbing actions.

You were probably hoping for lighter, more escapist reading from my summer list; but if you can settle instead for an important and enlightening narrative about a serious problem that needs addressing in the health and healthcare space, then I highly recommend this book.

Podcast Recommendations

Every evening in my house, right around dinnertime, I open the conversation with “I heard this really interesting podcast today…”

I love podcasts because when they’re done well, they are wildly interesting, informative, and entertaining. And I get to hear people talk on subjects I would never have access to in any other way.

Like all of you, I am often just too busy to sit and read as much as I would like. I can listen to a podcast while I’m driving or flying, walking the dogs or cooking dinner — or when I can’t sleep at 2:00 AM. You get the idea.

The Design of Business

This podcast is recorded at the Yale School of Management and is about how design works within complex organizations to shape decisions, products, and more.

Guests include clients from many industries and designers in many fields, and I absolutely love learning about the challenges and successes of how design, when embraced fully, has the power to improve all aspects of the work teams do, and the products and services they deliver.

Here’s an episode I especially like about a new design for an insulin kit for children with Type 1 diabetes:

And another with Somi Kim, Senior Director of Healthcare Solutions at Johnson & Johnson. I LOVE the way Kim describes her work: healthcare, she says, is about more than just products and procedures. It’s about observing people and learning about their needs. To be a designer now, one has to consider the breadth of an experience and to prioritize what will make the best difference for that person — someone who isn’t, at heart, a patient or a healthcare provider, but a human being.

We can’t, Kim says, do this in a vacuum, alone; we need to partner with many, including those whom we want to benefit from our products and solutions.

And there you have them: my annual hammock reading and listening recommendations. And hey: if you’re in the neighborhood, stop by! There’s most likely a watermelon mojito here with your name on it.

Posted in Books, Newsletters | Leave a comment

What the Hex?

I think (and hope) that I’m exceptionally open to new ideas and to change — but if I’m brutally honest, I must also acknowledge that there are times when my inner skeptic whispers (OK, shouts), “change is good — you go first!”

Maybe there’s something in my Myers Briggs Type (I’m an ENFJ) that can explain this tendency (or maybe not); regardless, I recognize that I often need to research, discuss, and think about a new concept before I’m ready to embrace it. Will I decide, “that’s a knuckleheaded idea, and hell no, I won’t go”? Or will it be, “you idiot — why didn’t you think of that?”?

Case in point: the tried and true (and my favorite for displaying geospatial data) Choropleth Map vs. the new (and not my favorite, though rather sexy) Hex-Tile Map.

Let’s start with the Choropleth Map, and why it works so well. A Choropleth (from the Greek words for “area/region” and “multitude”) is a thematic map in which areas are shaded or patterned in proportion to the measurement of the statistical variable being displayed, such as population density, or in the example below, rates of uninsured adults in the U.S.

Choropleth Maps provide an easy way to visualize how a measurement varies across a geographic area, or to show the level of variability within a region. Dark saturations of a color represent higher values; lower saturations, lower ones.

In the following display of uninsured adults (captured by the 2016-2017 U.S. Census Bureau, American Community Survey) we can see from the darkness of the blue that Texas has the highest uninsured rate (about 17%), while the lightest saturation of blue is in Massachusetts, where we enjoy an uninsured rate of about 2.7%.

(If we wanted to show the exact rates captured by the survey, we could add those numbers to the map. For the purposes of this newsletter, I will let the saturation of colors supported by the key speak for itself.)

The power of this display is that it allows the viewer to quickly and easily see and understand where the higher and lower rates of uninsured adults are across the country — to grasp and process the data at a glance. I also like being able to see each state in a way that is familiar to me: that is, each one has its actual map shape, showing its size, borders, and relation to other states.

Next, let’s consider the genesis of the Hex-Tile Map, as well as those few instances when I can accept that it may be useful. Most often associated with game-board design (remember Dungeons and Dragons?), a Hex-Tile Map (short for the six-sided hexagon shape) has the advantage of all its shapes being identical in size, eliminating any potential incorrect perception about the proportions of the information being displayed on a traditional geographic map.

The Hex-Tile Map helps us reduce or eliminate what is sometimes referred to as “the Alaska Effect”: the distraction or misperception caused by Alaska’s being really big (twice the size of Texas), and far away from the lower 48.

Further, some will argue that because states in the eastern U.S., especially those along the Northeast seaboard, are typically far smaller than those in the western section, it harder to see those states’ shapes, and read|analyze the values being displayed. Here is the same data displayed above in a Choropleth Map presented in a Hex-Tile Map.

The advantages I’ve cited of the newer type of map haven’t so far persuaded me to trade in my Choropleth for a Hex-Tile when presenting this type of rate data. I must admit however that Hex-Tiles sometimes offer a good medium in which to display simple “yes/no” data, and give designs a fresh, modern feel.

Consider the display below of states that have and have not expanded Medicaid under the Affordable Care Act (ACA): there’s no doubt that it’s very easy to see each state’s status, and to consider and analyze the information quickly and easily.

Additionally, when made interactive to serve as a filtering or navigation tool in a multi-faceted display, Hex-Tiles provide a nice alternative to drop-down lists and radio buttons.

Choropleth maps were a complete game-changer when they were used by Dr. Jack Wennberg to show unexplained variations in healthcare delivery in the U.S. (Take a look at The Dartmouth Atlas of Health Care). Their impact has been profound, raising our awareness about the disparities and variations of care delivery in the U.S. system, and moving people to question the status quo. I found them so thoroughly captivating that I quit my job and went back to graduate school for a second time, changing the focus and trajectory of my career (talk about change!).

I may be (slowly) coming around to the value and effectiveness of Hex-Tile Maps under certain limited circumstances, and at the end of the day I do see a few simple uses for them.

And what the Hex — a little bit of change never hurt anyone… much.

Posted in Communicating Data to the Public, Data Visualization, Newsletters | Leave a comment

A New View of Scores and Percentiles

Do you remember taking standardized tests? The one I remember best was the SAT’s, in high school. They struck fear into our hearts: would my friends and I do well enough to get into the colleges of our dreams?

This uncertainty made us worry (a lot) about our raw scores; but we obsessed just as much — or more — about how well (or poorly) we did compared to everyone else. Where did we rank alongside the competition? How would the ranking affect our chance at those coveted college spots? We were fixated on percentile results.

Here’s why. A percentile (as opposed to a score) tells you how you did compared with everyone else.

For example, if you were in the 90th percentile on the math sections of the SAT, you’d scored higher than 90% of the students who took the test with you. 47th percentile meant better than 47% of those students, and so on. Percentile results often made the difference between our first and second — or dead, desperate last — college choices.

Note further that a percentile score is not like a grade out of 100. If you’re in the 90th percentile, this doesn’t mean you got 90% of the questions right. It just means that compared with everyone else who took the test on that day, you scored higher than 90% of them.

Fast forward to survey results and comparisons that we encounter today, such as the Consumer Assessment of Healthcare Providers and Systems (CAHPS). These results may be displayed with the additional context of comparison percentile results; the big challenge is how to get all of this information into a visual display easily understood by the viewer.

Here’s an approach worth considering.

Survey Results

The second of the two graphics below displays how a specific group scored on several different survey questions (e.g., “Did you receive instructions about your medications?”). The display includes comparisons to all survey results.

The following diagram explains each part of the graph below it.

(click to expand)

(click to expand)

Score Scale

Now we can see (for example) that on Question 2, the group scored 90 (blue point), which was higher than the results from the 25th to the 75th percentile (gray bar), and than the 50th percentile (black line); and just slightly lower than the 90th percentile (orange line) results for all other groups combined.

The Price of Blood Tests

Interestingly, not long after I started writing this post, the following display of the price range for of one of the most common blood tests in medicine (metabolic blood panel) in different U.S. cities was published in the New York Times.

(click to expand)

In this display, the overall approach is fundamentally the same as in the first example I displayed above. A bar is used to show the range of prices from the 10th to the 90th percentile, or the low and high prices in each city, and a point to show the median price. The display also includes the price details on each bar, making it easy to see and compare the prices between cities.

Final Suggestion

If you feel a bit confused by the percentages and percentiles in the second graphic, versus your understanding of the somewhat clearer display of dollars and percentiles in the third, you’re not alone — but the former is the data we work with regularly, so we need to find ways to communicate it effectively.

If you think this type of display will work for some of the data you have (and I encourage you to give it a try), then I suggest that you create and include a diagram like the one in the first graphic — the “Score Key” — which explains the different parts of the display to viewers who may be new to your data, or who may have little experience with statistics.

In full disclosure, I found myself struggling to explain the first display quickly and easily before I sat down and studied it a bit. But once I saw the Score Key (thank you, Lindsay, on team HDV), it became crystal clear to me.

Hey, we’re all in this together and we all need a little help once in a while — well, that, a cocktail, and a nap.

Posted in Communicating Data to the Public, Data Visualization, Newsletters | Leave a comment

Leading Causes of Death

It’s a beautiful snowy morning in New England. There’s a fire in the fireplace, and my husband is reading the newspaper with two pups snoozing away at his feet. It’s all quite lovely and picturesque.

Well, it would be if it weren’t for the fact that I’ve been contemplating displays of data on the leading causes of death. Nothing like a table of death statistics to ruin a perfectly serene morning.

Here’s how it all started.

Cambria Brown, one of our HealthDataViz team members, sent me a link to the Pan American Health Organization’s website with a note that she was itching to do over its displays. I took a look and agreed: there is most assuredly room for improvement. The web page displays its data in three different formats, to all of which we have offered changes and corrections.


Current (Below):

The first display is a color-coded table, where the columns show the top five causes of death, encoded with different colors, and the rows represent age strata. The table allows the user to choose the country, age, time-frame and gender displayed.

(click to expand)

Proposed (Below):

We kept the original layout of the rows and columns, which allows the viewer to see and compare the leading causes of death by age; but made the following changes.


  • selected a broader range of colors, so that the different causes of death are easier to spot.
  • added an interactive slider that lets the user select up to ten causes of death versus only five in the original table.
  • displayed the rates per 1,000 people in each square of the table, providing useful additional context.
  • eliminated the Key, since everything in it is already in the table, making the former redundant.

(click to expand)

(click to make interactive)


Current (Below):

The second display on the site is just plain wrong. As I explained in my February 2019 newsletter,

  • Treemaps present data hierarchies not easily displayed via a bar graph.
  • The data here is neither hierarchical nor complex, and is clearer displayed as a bar graph.

(click to expand)

Proposed (Below):

If the goal is to display only one data dimension, stick to a bar graph. If however you want to show several dimensions of hierarchal data, an interactive Treemap works well, highlighting

  • each country and its size (the largest squares outlined in white).
  • the leading (top ten, for example) causes of death in each country.
  • how each cause has changed from one date to another (different saturations of blue for decreased and of orange for increased).

Note: a display like this one works only if it is interactive, so that the user can hover a pointer over each measure to see the information that cannot fit completely and be fully shown in certain areas (such as the white boxes) of the display.

(click to expand)

(click to make interactive)


Current (Below):

This image of small multiple maps, the site’s third display, also needs some work. In the one shown, “All Causes of Death” has been selected.

The display is flawed in that

  • although “(All) [Causes]” has been selected, only the top four are shown — and those are in alphabetical order, not in numerical order by frequency per country.
  • This means that if the viewer is hoping to see the leading causes of death in each country, the format utterly fails to deliver.

(click to expand)

While a user can select a specific cause of death to generate a map, there is still no quick, easy way to determine the #1 cause of death in each of the countries for a given year (or years). In the view below, the user would have to

  • select each leading cause of death, then hold that value in short-term memory (impossible: there are 67 causes) until the largest cause for the country of interest finally pops up.
  • A user wanting to know the leading cause of death in each country displayed on the map would have to generate 67 different views of the map for each one to make that top cause display.

(click to expand)

Proposed (Below):

Now consider the view Cambria created. This map displays

  • the top cause of death in each country and territory for the year[s] selected, displayed in a bar chart clearly organized by Leading Cause, Country and Rate per 1,000 People.
  • All information is visible and graspable in a single compact, revelatory view.

(click to expand)

(click to make interactive)

At the top of her display, Cambria has placed a button to call up trends for the leading cause of death by country, and show how those trends have evolved over time.

This view clearly displays

  • for all countries the top ranked causes of death beginning in 2001, how each ranking changed, and where it stood for the last year selected, 2013.

For example, in 2001, Dementia|Alzheimer’s was the 11th leading cause of death worldwide; by 2013, it had risen to 5th — a perception both useful and sobering.

(click to expand)

(click to make interactive)

Final Thought

Here’s a tip about all of this.

A few simple case studies of what you believe that your displays’ end-users should be able to see and understand are extraordinarily useful during user-experience (UX) testing.

Everything I have said here could have been discovered via this simple technique, and would have driven crucial, transformative design changes. I have written in previous newsletters about the central importance of understanding why displays like Treemaps were created, so that you use them effectively.

Well, the snow has stopped falling and it’s time to find my way to a shovel while I imagine how beautiful my daffodils will be in a month. Hope — for splendid flowers and good design — really does “spring” eternal!

Posted in Communicating Data to the Public, Data Visualization, Newsletters, Tableau, Tables | Leave a comment

Mekko Charts

I’m a long (long)-time devotee of the iconic Marimekko designs and textiles. I have Marimekko clothes (I’m sure they’re considered vintage now), and in my daughter’s bedroom I splurged on yards and yards of custom drapes made from Marimekko’s Pieni Unikko Blue fabric.

They are big and bold, and make a huge first — and lasting — impression. My family can bury me with those drapes, and I will pass into my next life a very happy woman (I kid you not).

Some Marimekko designs are geometric, like this one (below), called Ostjakki Print — and yes, you guessed it, they have been the inspiration for a type of data display called a Mekko Chart.

At the risk of being a killjoy YET AGAIN, I will say that Mekko charts can be fun to look at and play around with; but as evolving research is helping us figure out, what we like and what is best for allowing us to examine and understand data are often two entirely different things. Sort of like the candy versus carrot conundrum.

Below is an example of an interactive Mekko chart. It was created by Josh Cothran (Georgia Institute of Technology) for the California Health Care Foundation (CHCF) to display U.S. healthcare expenditures:

(click to expand)

This Mekko chart illustrates healthcare expenditures by two variables: service and payor.

Each row shows a type of service (hospital care, prescription drugs), and each colored section of the bar represents a payor (Medicare, Private). The height of the row represents how large that particular service type is compared to all other services being displayed; the width of the different colored sections represents each payor’s portion of that service type.

You can get a general sense of the different service values in this display, and of which ones are larger or smaller, through the height of the rows (labels provide the specifics), but it’s hard to compare these heights directly.

Services aren’t ranked, making it difficult to consider the magnitude of spending in an ordered way. It’s also just about impossible to understand and compare the different colored sections of the bars, because they begin and end at different places on the scale.

And although you can hover over a section to view details, it is virtually impossible to hold all the information presented in short-term memory and make comprehensible comparisons; we simply aren’t wired to do this (which is why we use a technique called “ chunking” to help us remember things like phone and Social Security numbers).

Now consider a different display of this information that I collaborated on with our newest HDV team member, Lindsay Betzendahl (thank you, Lindsay!).

2017 us healthcare expenditures

(click to make interactive)

In this new display of the data, we have used a simple heat map to directly display the expenditures by both service type and payor, with the darker saturation of blue representing the higher values.

Additionally, the rows are used to rank the expenditures from high to low, based on the overall percentage of service-type expenditures, which is also clearly displayed in the accompanying bar chart on the right side of the table.

More contextual information is displayed in an accompanying set of figures in the far right column, which shows the percentage of change in the amount spent in each category from 2011 to 2017.

A similar treatment is used at the bottom of the table for payor type, where you can view the expenditures by each payor, their proportion to the total, and how this relationship changed from 2011 to 2017.

Now we can easily see and understand all the information being displayed.

For example, in 2017, the largest expenditures (more than $1 trillion) were for hospital services at 38.6%, and Private Insurers paid the largest share, at $455 billion. We can also see that all hospital expenditures increased from 2011 to 2017 by 34.1%.

Interestingly (but not surprisingly, given the expansion of Medicaid under the Affordable Care Act|ACA), we can see from the figures at the bottom of the display that the largest change in payor expenditures from 2011 to 2017 was attributable to an increase in Medicaid spending of 39.4%.

Cool graphs are indeed very cool, and I understand their ability to wow an audience. But when we step back and think critically about what we can easily comprehend and explain while viewing them, versus what we can understand and convey via less dazzling displays of the same data, we find (and remember, there is emerging research to support our discovery) that the oft- misunderstood and dismissed wall-flower of displays is the most effective, time and time again.

And remember, folks, at the end of the day it is our job to make health and healthcare data clear, comprehensible, and compelling — even if it doesn’t look as handsome as a set of vintage Marimekko sheets.

Posted in Best Practices, Communicating Data to the Public, Data Visualization, Newsletters, Tableau | Leave a comment

Design Thinking Required

A colleague recently sent me an article by Atul Gawande, MD: “Why Doctors Hate Their Computers” (huh — you don’t say!).

If you’re familiar with Dr. Gawande’s work, you know that his articles and books are both thoughtful and thought-provoking (he’s a multi-award-winning author and a MacArthur Fellow for good reason). I read the article with interest, recognizing many of the frustrations with new Electronic Health Records (EHR’s) noted by the physicians he interviewed.

But then there was this pronouncement by one Chief Clinical Officer in response to clinicians’ frustrations: “[W]e think of this as a system for us and it’s not,” he said. “It’s for the patients.” I was struck speechless (which also means that hell may actually freeze over).

Hold up! EHR’s are and should be for use by, and the benefit of, BOTH physicians and patients.

And let’s be clear: there’s no “which came first” chicken|egg conundrum here. Aside from some electronic portals that allow patients to enter their personal and medical history, the majority of EHR data and information comes directly from clinicians and other providers caring for patients.

Therefore, patients will realize the full benefit of the new EHR’s only if they are intuitive and easy to use by clinicians — that is, only if clinicians are able to use them optimally.

Clearly, we need EHR’s for all of the reasons noted in Gawande’s article, and I fully understand the frustration the Chief Clinical Officer and others feel when they introduce new technology to a sometimes hostile audience. (I am after all a charter member of the “Innovators who are Deeply Maligned and Misunderstood” club.)

However, this particular comment by the CCO causes me great consternation and reinforces my conviction that Design Thinking methodologies are not well understood or being used to create better systems for ALL stakeholders.

Rather, and all too often, the methodology of Design Thinking is missing entirely. (Sadly, any reference to it is also missing from the Gawande piece.) Instead, teams with no training or experience in Design Thinking develop systems using mostly traditional engineering approaches. Then, when users find these systems difficult and frustrating to use, the response tends to be:

bang head here and carry on

Design Thinking is a process for creative problem-solving that puts people at the center of the design process through empathy. It is human-centered, collaborative, experimental, and optimistic. The process is not linear, but usually follows these steps:

Empathize. Empathy in Design Thinking requires you to observe the people you are designing for, interact with them to understand their points of view, and immerse yourself, so that you can experience what they experience. It helps move teams away from self-referential thinking. (Check out my post about personas and how they help us empathize here.)

Define. As you define a problem in Design Thinking, you are looking for interesting and compelling insights about the people you are designing for, and how they think about the work they do. This search provides focus and a framework for your problem-solving efforts. It allows you to be intentional in your designs. (Check out my post about mental models here.)

Ideate. By building lots of design solutions, you allow for unexpected and radical solutions to design challenges. You also harness the collective perspectives and strengths of your design team. (Check out my post about the power of sketching ideas here.)

A Prototype is a way to test for functionality, and permits further insight from the people you are designing for. Prototyping allows you to fail quickly and cheaply by piloting ideas before fully implementing them. (Check out my must-read recommendation, a piece by Alan Cooper on interactive interface design, here.)

Test. By testing your ideas with actual people, you can both refine the concepts and learn more about the people—their needs and desires. Testing also allows you to examine whether or not your design solution is solving your original design problem, or if you need to re-examine your prototypes. (I highly recommend Steve Krug’sbooks on usability testing.)

In closing, I think about Cooper’s first book, The Inmates are Running the Asylum,which should be required reading for everyone involved in creating these systems.

Cooper argues convincingly that designing interactive software-based systems is a specialty as demanding as the construction of them. He forcefully and correctly asserts (and his assertions are further borne out by Dr. Gawande’s article) that the cost of bad design is incalculable, as it robs us of time, customer loyalty, competitive advantage, and opportunity.

Bottom line: it is long past time to stop blaming system users by labeling them “disgruntled” and “uncooperative,” loftily declaring that they simply “need to get with the program.”

Instead, we must get our collective act together and use Design Thinking methodologies to create systems that humans love to use.

Posted in Communicating Data to the Public, Know Your Audience, Newsletters | Leave a comment

A Profoundly Moving Data Display, Revisited

Perhaps you’ve visited this memorial. There are more than 58,000 names engraved on panels of polished black granite commemorating the Americans who died or were listed as Missing in Action in the war. The 250-foot long walls are each ten feet tall at their apex and gradually slope down to ground level. Viewers see their own reflections in the stone as they read the names inscribed there.

The obvious (and perhaps most neutral) way to list the names would have been alphabetically by last name. Instead, designer Maya Ying Lin chose to list them chronologically by date of death (or day reported missing).

Ordering the soldiers by date of death serves to place them near one another as they may have fallen on the battlefield. It helps other soldiers who served at the same time remember those whose deaths occurred during their own tour of duty. It encourages visitors to contemplate the sacrifice of each soldier, and to wonder at the connection of other visitors to the memorial.

The simple, beautiful, and brilliant design of this memorial is really something quite extraordinary in its dignified and engaging presentation of seemingly straightforward information — the names of soldiers.

The Vietnam Veterans Memorial is a profound example of meaningful data visualization, and of the importance of design in communicating the message in our statistics. Alphabetically, the names are just that — data. When listed chronologically, as here, the same information tells a deeply moving story.

The Vietnam Veterans Memorial gets it exactly right on all levels. It weaves a narrative that draws viewers in, connecting them to one another, and both leading and permitting them to reflect on their feelings about the war. It is a prism through which memories of and thoughts on our most controversial and divisive conflict may forever change as it re-molds the most firmly held beliefs — raising awareness, suggesting answers, perhaps stimulating new questions. It can even inspire action, changing what people do in the voting booth, in their career choices, in their communities or as volunteers.

The Vietnam Veterans Memorial sets the bar pretty high for the rest of us, and that is a good thing, because in doing so, it reminds us that even in our day-to-day work of reporting healthcare data, the way we do that really matters.

It matters because we need people to pay attention to important information, and to engage with it — to change how they think and work because of what we have shown them. Data presentation (on black stone or white board) matters because the only way we can affect our systems of care is to be moved to action by what the presentation reveals — to action that will make those systems, and the people they serve, better.

Posted in Communicating Data to the Public, Data Visualization, Know Your Audience, Newsletters | Leave a comment

Serious Talk About Bubbles

This past summer our beloved daughter, Annie, was married to our new, very favorite son-in-law, Douglas. There were bubbles blown by the guests as the bride and groom finalized their vows, and flowing Champagne (lots and lots of Champagne). It was everything we hoped for in every way imaginable. Quite simply, in the words of our sweet girl, “the day was perfect.” Indeed it was.

Now, dear reader, you know where this is going… all those bubbles got me thinking about the bubble charts I see in my work with clients. Alas, unlike Annie and Doug’s wedding, they’re not so perfect.

Consider the following example that was on a Hospital Report Card I recently received:

Click to expand

My initial reaction upon viewing this display was “WHYYYY???? (imagine my whining tone for emphasis).

Why would anyone display data in this manner?

I have a couple of guesses: it’s eye catching and “fun.” The software lets you do it, so it must be right. While those may be real factors governing the creation of this display,

that doesn’t mean they’re acceptable, or in accordance with data visualization best practices. Here are a few reasons why:

  • Not all of the bubbles are clearly labeled with the category of data being displayed, nor could they be, given the size and spacing of each one.
  • Even if the first problem could be remedied by the addition of a color-coded key, the key would be so long that no mere human could ever hold all of its information in short-term memory while viewing the display.
  • The colors are certainly bright and shiny, but they’re also distracting, and add no value to the viewer’s overall understanding of the data.
  • The value of each category of data is not labeled, and it is difficult (if not impossible) to make direct comparisons between one category value and another.
  • Categories can’t be ranked or ordered in any logical way.

As always, our first and over-arching objective must be to show the data and the story in it.

Enter the far less showy and oh-so-sensible, ever-practical bar chart. Displaying data using a bar chart affords us the ability to show the entire label for each value. Additionally, with the use of only one color, the viewer is no longer distracted by trying to understand what the different colors mean (nothing), and instead can see the shape of the data. It is also possible to directly label the value of each bar being displayed, and to rank the results or display them in some other potentially meaningful way, such as alphabetically by category.

Click to expand

Click to expand

Now let’s consider another scenario where bubbles hinder our ability to show the data clearly and in context.

Imagine that we have been asked to create a display for a provider group that delivers services for patients (male and female) diagnosed with reproductive issues. The display needs to include the number of cases in each category for:

  • the current month,
  • the year to date compared with the previous year to date and the difference, and
  • a twelve-month rolling trend.

For this example, let’s also assume that we are displaying data by calendar year, and that the last month of available data is for June of the current year.

Click to expand

Employing these techniques, we can use our label once and display the current month’s case count, followed by year-to-date versus prior year-to-date, and the 12-month rolling trend. Now the viewer can easily see that there are more cases for male than female in the current and year-to-date data, and that this seems to be an ongoing trend.

I know, I know: bars are boring; bubbles are fun. But that is not the point. The goal here — always — is to convey the data in a clear and compelling manner that will make the story in it stand out, and move people to inquire further, learn something new, and when appropriate take action.

If all the same you’re really feeling the need for some fun bubbles, I’ve got a case of leftover Champagne that I’d be thrilled to share.

Posted in Communicating Data to the Public, Dashboards, Data Visualization, Newsletters | Leave a comment